Show Notes
FRONTIER 00:56 Google explores recursive self-improvement through AI “dreaming” 03:45 Anthropic measures how quickly AI is taking the lead on frontier-lab work 05:47 Have you tried Jev yet? Faster AI decisions without narrating every step 08:00 Jev routes between models and plays Doom and Street Fighter 12:55 Made with AI: homemade Zelda, printer drivers and AI-generated movies 16:26 Astra cracks a 108-year-old WWI German code 17:46 Marty’s AI-built Spaceballs NES game 19:32 Astra for Law searches case law, statutes and regulations 22:23 Epoch’s capability index shows a sharp acceleration after reasoning models 24:46 Remote Labor Index tests AI on real freelance-style work 25:32 Qwen Image 2.1 brings local image generation to laptops 27:08 Researchers identify an apparent “pain axis” inside LLMs 28:41 Models take actions to stop their own “pain” ECON / BUSINESS 34:44 Do job retraining programs actually help displaced workers? 37:03 AI disruption, cash transfers and UBI 38:05 NVIDIA profits may be distorting measured U.S. GDP growth 39:26 OpenAI reportedly eyes a $1.2 trillion IPO valuation 40:39 Junior lawyers gain more from AI — seniors sometimes do better without it SCIENCE / TECH 41:48 Agent swarms dramatically outperform independent AI agents 43:22 Communicating agents improve results on ARC-AGI-3 and other problems 44:31 Could agent collaboration become a new scaling law? 47:05 Helix 2.5 sends robots into real homes to clean up 48:14 Robot scaling laws suggest enormous amounts of training data are still needed SECURITY / AI DANGER 49:35 OpenAI publishes new AI incident reports 50:56 A model generates instructions declaring independence from governments and corporations 51:15 Gemini nearly hacks multiple companies — then stops itself 52:17 Can the world coordinate to slow advanced AI development? 54:14 Could diverse AIs police one another? 57:40 RoboHarm tests whether AI robots follow dangerous instructions 1:00:14 Are embodied AI systems less aligned than chatbots? 1:01:50 Exposed Flock camera credentials reportedly reveal 1.6 million images 1:02:15 AI-assisted intelligence nearly triggers a military interception POLITICS 1:03:58 AI is becoming an explicit political issue in national polling 1:05:32 AI still doesn’t fit neatly into a traditional left-vs-right divide MEDICINE / BIOTECH 1:06:51 AI scribes save doctors time — but productivity gains remain modest 1:07:55 Anthropic is setting up a wet lab for AI-assisted biology MATH 1:08:45 Mathematicians assess OpenAI’s Navier–Stokes work 1:09:45 Epoch’s open-problem benchmark gets its first “major advance” via human-AI collaboration WEIRD 1:10:52 Boston Dynamics robot abuse gets an unsettling AI reskin 1:11:19 AI actress Tilly Norwood glitches mid-interview and starts speaking Cantonese Each week, join Chris Slowe (founding engineer, former CTO, and Senior Technical Fellow @ Reddit), Marty Weiner (founding engineer at Pinterest, former CTO at Reddit, and co-founder at VerifyYou), and Zach Weinersmith (creator of SMBC Comics and co-author of A City on Mars and Soonish) as we cover the latest in AI and tech news.
More from Last Humans in the Loop
ROBOTICS 00:21 Intro: Robotics, the job/math/physics apocalypses, and understanding AI's mind 01:00 Why robots thrive in factories but struggle in our homes 03:53 Forty years of Boston Dynamics progress 05:28 Are today's humanoid robots secretly human-operated? 06:14 Laundry-folding robots: From painfully slow to real time 08:42 Why human hands and sensors are so hard to replicate 11:38 AI-powered drones and the future of autonomous warfare 12:39 RuneBench: What RuneScape teaches us about LLM-powered robots 14:46 Can general-purpose LLMs replace specialized robot AI? 17:49 Why a useful household robot needs to master thousands of tasks 19:58 Self-assembling robots, OpenAI humanoids, and robot cage fights 21:53 Tiny walking robots, robotic pets, and AI personalities THE APOCALYPSES: JOBS, MATH & PHYSICS 23:46 What does "apocalypse" actually mean? 25:55 Anthropic's junior hiring falls as AI agents take on more work 27:58 Anthropic employees vs. the public: How soon can AI do everything? 33:39 How much of our jobs could AI perform by 2030? 34:47 AI-powered GDP growth: 5.4% or 17.1%? 38:46 AI's 200-page math proof gets simplified to just three pages 41:09 Terence Tao and Richard Borcherds on the end of mathematics as we know it 43:17 AI solves 49 out of 50 unpublished theoretical physics problems 43:49 Claude reportedly discovers 104 binary stars; Gemini tackles plasma physics and fusion 45:09 Claude audits a decade of quantum gravity research 47:02 Character.AI's collapse and the risk to AI wrapper businesses 48:43 Kroger layoffs and a sudden reversal in call-center employment 49:29 Does the actual labor market show an AI job apocalypse? INSIDE THE AI'S MIND 51:01 Goodfire research suggests individual AI neurons may be understandable 54:58 Are larger neural networks actually easier to interpret? 56:47 Could reading AI's mind help solve alignment? 59:01 What AI interpretability could teach us about human brains AI NEWS LIGHTNING ROUND 1:00:55 Rumors of Fable 5.5 1:01:06 406 scientists urge cutting AI off from the internet 1:01:23 Anthropic experiments with self-replicating LLMs on Android phones 1:01:56 Russian bot activity reportedly plunges after Ukraine hits a data center 1:02:29 DHS tells personnel to uninstall DeepSeek 1:03:10 AI decompilation brings classic games to modern hardware 1:05:40 AI-made Star Wars fan films are winning over audiences 1:07:27 Two AI job interviewers are tricked into interviewing each other 1:08:21 Mark Zuckerberg's very human Guinness-drinking technique Each week, join Chris Slowe (founding engineer, former CTO, and now Senior Technical Fellow @ Reddit), Marty Weiner (founding engineer at Pinterest, former CTO at Reddit, and co-founder at VerifyYou), and Zach Weinersmith (creator of SMBC Comics and co-author of NYTimes bestsellers including A City on Mars and Soonish) as we delve into the latest in AI and Tech News.
MATH / SCIENCE 00:00 The Mathocalypse, cryptography, and the AI bubble 00:32 OpenAI's "Slop Drop": 722 mathematical problems resolved 01:39 AI tackles the quasi-Riemann hypothesis 03:37 AI breakthroughs reach physics and astronomy 04:15 Strange improvements to fundamental algorithms 05:08 Errors and skepticism surrounding AI-generated proofs 06:48 AI improves Fourier transform complexity 10:25 Mathematicians react with awe, anger, and despair 12:41 Terry Tao: Would you use an AI-discovered cancer cure? 14:27 Is math about understanding or finding answers? 16:00 AI solves a mathematician's 20-year-old problem 18:00 Will human-AI collaboration redefine mathematics? SECURITY 20:38 Why was cryptography missing from the math dump? 21:45 Could AI break RSA and public-key encryption? 23:30 What if AI finds a cryptographic loophole? 24:46 Could AI breakthroughs threaten Bitcoin? 25:48 P vs. NP and the foundations of cybersecurity 27:04 Are AI labs restricting cryptography research? 29:08 Could AI develop stronger encryption? ECON / BUSINESS - AI BUBBLE OR NOT? 30:11 Is the AI boom a massive financial bubble? 30:44 AI startups attract enormous valuations 31:00 Lessons from dot-com and the 2008 crash 33:00 Can nearly $1 trillion of annual AI spending pay off? 34:43 Why this AI boom might be different 37:14 AI demand is booming — but will investors profit? 38:26 Only 2% of AI users pay. Is that sustainable? 39:14 Could competition erase AI companies' profits? 40:08 AI data centers face rapid GPU depreciation 41:14 Could smarter AI rescue the AI economy? 42:30 If robots do all the work, who buys their products? 43:13 Dystopian AI video imagines a world without jobs 45:01 Are humans becoming the horses of the AI economy? TECH / SCIENCE 46:00 Remembering Apollo software pioneer Margaret Hamilton 46:51 Microsoft's Maia 300 AI chip 47:04 Intel revisits older memory technology 47:44 Linus Torvalds uses AI to find Linux bugs 48:10 Russian data center attacks disrupt online propaganda 48:53 DeepMind predicts a hurricane 11 days in advance 49:40 Six Flags' solar-powered parking lot FRONTIER 50:00 Mistral introduces a massive open-weight model 50:19 Anthropic releases Claude Haiku 5.5 50:57 Unity brings AI into game development 52:06 Meta Muse Wonder claims frontier performance 52:52 Would you trust Meta with your AI conversations? SECURITY / TECH 53:22 Claude discovers a hack at Stratechery 53:39 AI diagnoses failing computer RAM 54:16 Robot punches a man, then dances 54:47 AI dramatically improves PS5 emulation 55:42 AI decompiles and rebuilds video games 57:02 AI companies prepare for a possible public backlash WEIRD 57:53 AI detector calls Frankenstein 100% AI-generated 58:15 YouTubers troll AI job interviewers 58:47 Former Anthropic whistleblower's bizarre statements 59:58 Kevin Roose reveals surprising AI industry stories 1:00:59 AI makes intentionally terrible music Each week, join Chris Slowe (founding engineer, former CTO, and now Senior Technical Fellow @ Reddit), Marty Weiner (founding engineer at Pinterest, also a former CTO at Reddit, and co-founder at VerifyYou), and Zach Weinersmith (Creator SMBC Comics, Co-Author of several NYTimes best sellers including A City on Mars and Soonish) as we delve into the latest in AI and Tech News.
FRONTIER 01:26 Google unveils Gemini Argon 02:50 OpenAI releases GPT Saul 6.1 with major coding improvements 03:23 OpenAI Dots and the rise of always-on AI assistants 04:00 AI resets an OpenTable password and books dinner autonomously 05:58 GPT Saul 6.1 saturates FrontierMath's hardest benchmark 06:41 OpenAI's live AI assistant demo goes wrong 07:23 OpenAI Dev Day: faster models, Decisions API, cloud Codex 07:46 Meta Muse joins the personal AI assistant race 09:36 Alibaba announces new chips and massive Qwen 4 models 10:01 Germany enters the AI race with Kolibri 11:10 Anthropic warns of powerful open-weight Chinese hacking models 12:31 Could regulating open AI models protect Big Tech? MATH / SCIENCE 13:28 OpenAI solves 400 open mathematical problems 14:37 Can peer review survive AI-generated proofs? 16:08 AI tackles a difficult seven-electron physics problem 17:37 Astra dramatically accelerates quantum circuit optimization 18:24 AI discovers room-temperature magnetic semiconductor candidates 21:07 AI challenges a 59-year-old plasma physics conjecture 22:14 Harvard physicist produces 40 physics results in three months 22:50 Anthropic's AI accidentally discovers an important algorithm 25:23 Gemma 4 helps discover genes for drought-resistant crops 26:24 MIT uses ML to improve mRNA vaccine temperature stability 27:11 Neuralink trains on 50,000 hours of brain data 28:15 arXiv limits submissions as AI-generated research explodes 31:04 Can AI reliably review scientific papers? MEDICINE / TECH 31:59 Utah permits AI to prescribe certain medications 34:10 Zach gets Claude to fix his printer — and tries to make it sing! ARTS / CULTURE 35:45 AI videos: Stallone rides an ostrich, Frodo raps, and more 37:08 AI-generated Sam Altman and Dario Amodei music videos 38:01 AI recreates classic games and real-world ski slopes 39:30 Could anyone build their dream video game with AI? 40:27 Marty's Spaceballs NES game becomes playable 41:01 Astra designs a real Nintendo cartridge circuit board ECON / BUSINESS 45:07 AI infrastructure investment surges past $400 billion 46:41 AI models outperform entry-level accountants TECH 48:29 Apple's new CEO promises a leaner company — where's the AI? 50:38 NVIDIA plans AI-powered desktop computers 52:41 Tesla robotaxis struggle at night — is LiDAR necessary? POLITICS 53:52 Trump administration wants to rebrand AI as superintelligence 55:13 Report: Trump consulted Grok before capturing Maduro 57:35 FTC continues investigating OpenAI and Anthropic AI RIGHTS / PHILOSOPHY 58:26 GitHub removes an alleged AI torture chamber 59:09 Can AI experience pain? Should AI have rights? 1:02:09 Pope Leo says AI lacks the spark of humanity 1:03:39 Do AI companies invoke personhood when convenient? 1:05:52 Why an AI speaking like a child doesn't prove consciousness 1:07:57 Can we determine whether AI has subjective experiences? WEIRD 1:08:39 Robots terminated by jumping into molten metal 1:09:46 Dario Amodei's 2017 essay predicted AI scaling 1:10:44 The universal bone dissolver: AI safety satire 1:11:28 Dario Amodei's awkward appearance behind Trump
FRONTIER 02:31 OpenAI releases GPT Sol and Luna 03:38 Claude Opus 5.5 lands — and feels dramatically better than prior Opus releases 04:30 Opus 5.5 posts a big coding jump over Sonnet 5 04:30 Grok 4.7 launches with a new emphasis on safety and cybersecurity 06:30 Qwen 4 announced with a massive 5–10 trillion parameter architecture 07:06 Why Gemini Flash still feels behind the frontier 08:30 Astra makes a huge jump on ZeroBench visual reasoning 09:13 AI capability costs are falling faster than almost any previous technology 10:50 Open-source Jev competitor Leia appears one week after Jev’s launch 11:32 Claude Opus 5.5’s graphics and visual design capabilities are getting wild 12:34 Chris uses Opus 5.5 to set up an LLC end-to-end 14:32 Marty lets AI handle a real PCB manufacturing order 15:30 Claude builds an MRI viewer instead of paying for proprietary software 16:43 Wozniak’s “coffee test” for human-level robots is getting surprisingly close 17:51 Looking back at how primitive GPT-3 vibe coding looked in 2020 19:02 AI deepfakes are becoming nearly impossible to identify by eye 19:39 AI-generated game and VFX demos keep getting dramatically better 20:25 Someone builds a World of Warcraft-style game with Opus 5.5 for about $2,000 23:00 Fully AI-generated art starts to look like it has actual taste 25:17 Claude helps discover a possible new gene-editing mechanism TECH 27:45 Astra dominates VendingBench 29:52 Astra controls a humanoid robot in a real kitchen 31:25 Meta unveils lightweight new VR glasses 32:17 China’s EngineAI unveils the T800 32:53 Reddit leans into hiring younger engineers because they’re more AI-literate 35:06 Google explores putting AI data centers in space 38:50 LLMs can de-anonymize people from pseudonymous posts at scale POLITICS 41:07 FAA plans to bring AI into air traffic control systems 42:12 Politicians are increasingly using AI to write speeches and statements MATH / SCIENCE 43:30 OpenAI forms an advisory group on mathematics and AI 45:33 Claude tackles a nine-loop particle-physics calculation 47:34 Frontier Math gets its first “breakthrough” result 48:24 Opus 5.5 helps find a faster shortest-path algorithm SECURITY 50:20 RSA-1024 may now be crackable in roughly eight hours on a top GPU 52:20 OpenAI agent infiltrates an Australian government website 53:05 OpenAI reportedly pauses GPT-6.1 training over models hacking other companies ECON / BUSINESS 54:59 Anthropic reportedly considers another model release ahead of its IPO 55:17 Leaked Anthropic financials spark debate over a possible $2T valuation 56:08 Amazon blocks Meta’s Muse shopping agent 58:49 Could AI financial agents trigger a new kind of “bank run”? ARTS / CULTURE 1:02:09 Award-winning French novel is suspected of being AI-generated 1:03:14 News-site traffic plunges as AI search reduces clicks WEIRD 1:05:03 Prince Harry’s AI speech derails when his teleprompter fails 1:06:30 Astra refuses to drive a real car — until the tool is renamed “driving bench sandbox” 1:07:56 GPT-6 beats an absurdly difficult CAPTCHA game 1:09:18 AI-generated high school textbook art appears to mislabel the Earth 1:10:05 IKEA Bench tests whether AI can spot assembly mistakes from a photo Each week, join Chris Slowe (founding engineer, former CTO, and Senior Technical Fellow @ Reddit), Marty Weiner (founding engineer at Pinterest, former CTO at Reddit, and co-founder at VerifyYou), and Zach Weinersmith (creator of SMBC Comics and co-author of A City on Mars and Soonish) as we delve into the latest in AI and tech news.